Plunger pump anomaly detection method and system based on pressure pulsation similarity

Abstract

The application provides a plunger pump anomaly detection method and system based on pressure pulsation similarity, comprising: collecting a pressure signal of a plunger pump outlet, removing a trend of the outlet pressure signal by using a trend removal algorithm, removing abnormal values caused by interference, performing Fourier transform, checking a frequency spectrum of the outlet pressure signal, normalizing the outlet pressure signal, calculating a health factor based on pressure pulsation similarity, comparing the health factor with a set threshold, and determining whether the plunger pump is abnormal. The application does not need a historical data set for model training, and can effectively perform state monitoring on the plunger pump.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plunger pump state monitoring, in particular to a plunger pump abnormality detection method and system based on pressure pulsation similarity. BACKGROUND

[0002] Plunger pumps are widely used in the industrial fields of engineering machinery, aerospace, robots, etc. due to their high power density, compact structure and high efficiency. The plunger pump is a key component of a hydraulic system, and the health state of the plunger pump directly affects the health state of the hydraulic system. For complex engineering machinery, effective and accurate online state detection of the machine can greatly facilitate the maintenance of the hydraulic system, save a large amount of downtime maintenance time, arrange spare parts in advance, save downtime maintenance costs, shorten the construction period, and improve the safety and reliability of the hydraulic system. For plunger pump health detection technology, many existing methods choose to install vibration or outlet pressure sensors to monitor the health state of the plunger pump. However, many current plunger pump state monitoring algorithms are based on supervised machine learning algorithms. Supervised machine learning algorithms require a large amount of historical health and fault data to train a machine learning model, and then use the trained machine learning model to diagnose faults or monitor the state of the plunger pump. However, it is very difficult to collect a large amount of plunger pump fault label data for actual engineering applications. Moreover, for a hydraulic system, the pressure of the hydraulic system changes with the load. Therefore, it is more difficult to collect fault label data under different working conditions.

[0003] Patent document CN106022388A (application number: CN201610366370.6) discloses a multi-feature fusion refueling pump abnormal working condition detection method. The method comprises the following steps: S1: using pressure sensors, liquid level sensors, potentiometers and other sensors to obtain refueling pump outlet (inlet) pressure, flow, outlet (inlet) temperature and current and other monitoring data; S2: using an adaptive filtering method to perform smoothing preprocessing on the data; S3: using a multi-core learning method to map the various heterogeneous monitoring data to a high-dimensional linear space, and fuse different feature components of the refueling pump monitoring data; and S4: comparing the fusion result with an adaptive dynamic threshold to identify the abnormal working condition of the refueling pump.

[0004] Therefore, the present application is based on this problem, proposes an unsupervised plunger pump anomaly monitoring method, without historical fault data, based on the similarity of pressure pulsation to implement the anomaly monitoring method of plunger pump, to cope with the problems of load change and difficulty in collecting a large number of fault label data in actual engineering and hydraulic system pressure. Due to the structure of the plunger pump (limited number of plungers, reverse flow, etc.), the outlet flow of the plunger pump is not a constant value, and changes with time pulsation, and the outlet pressure of the plunger pump in the healthy state is also a pulsation change time series, that is, a pseudo-periodic time series. In the healthy state, the health status of each part (for example: plunger) is similar, so the pressure pulsation waveform of each outlet pressure is similar. When a fault occurs (for example, the friction pair of the plunger pump leaks), the flow pulsation changes, and after the pipeline impedance conversion, the pressure signal also changes. Due to the change in the health status of part of the parts, the similarity of the pressure pulsation of the plunger pump in one revolution changes, so the similarity of the non-self-matching subsequence in the pressure signal of the plunger pump in one revolution can be monitored as the health factor of the plunger pump, and the anomaly monitoring of the plunger pump is realized by setting a threshold. SUMMARY

[0005] In view of the defects in the prior art, the present application aims to provide a plunger pump anomaly detection method and system based on pressure pulsation similarity.

[0006] The plunger pump anomaly detection method based on pressure pulsation similarity provided by the present application comprises:

[0007] Step S1: A high-frequency response pressure sensor is installed at the outlet of the plunger pump to collect the pressure signal of the plunger pump outlet;

[0008] Step S2: Determine whether the collected pressure signal has a trend item change, and if so, use a trend removal algorithm to remove the low-frequency change trend item;

[0009] Step S3: Remove abnormal points from the outlet pressure signal;

[0010] Step S4: Perform Fourier transform on the signal of the pressure sensor to obtain the frequency spectrum of the pressure signal, and determine whether the frequency with the highest amplitude of the frequency spectrum is the motion fundamental frequency or multiple of the plunger pump, if so, the plunger pump is preliminarily judged to be normal, and steps S5-S7 are executed, otherwise, the plunger pump is abnormal or the normal plunger pump pressure pulsation signal cannot be collected;

[0011] Step S5: Normalize the pressure signal;

[0012] Step S6: Calculate the plunger pump health factor based on the similarity of pressure pulsation for the normalized outlet pressure signal;

[0013] Step S7: According to the calculated health factor, it is judged whether the plunger pump is abnormal according to the preset threshold.

[0014] Preferably, the kinematic frequency of the plunger pump is calculated as follows:

[0015]

[0016] Wherein, n is the rotational speed of the plunger pump, and k is the number of plungers of the plunger pump.

[0017]

[0018] Wherein, FS is the sampling frequency.

[0019] According to the sampling frequency FS, the rotational speed of the plunger pump, and the number of plungers of the plunger pump, the data points L1 of the outlet pressure signal segment corresponding to a kinematic frequency f at the sampling frequency FS can be calculated:

[0020]

[0021] The data points of the outlet pressure points of the plunger pump rotating one revolution are L=L1*k.

[0022]

[0023] Preferably, the pressure signal is decomposed into k intrinsic mode functions IMFs and a residual term R by using a variational mode decomposition algorithm, and the expression is:

[0024] [IMFs, R] = vmd(T)

[0025] Wherein, vmd represents variational mode decomposition, T is the outlet pressure signal, IMFs is the intrinsic mode function, and R is the residual term.

[0026] Then, the IMFs are selected to find the IMFs where the trend items are located, and then the original signal is subtracted from all the trend items to obtain the detrended outlet pressure signal:

[0027]

[0028] Wherein, represents the detrended outlet pressure signal, represents the IMFs where the trend items are located.

[0029] Preferably, an outlier removal algorithm is used to remove outliers, including judging the difference between two points in the time series, and judging whether it is an outlier by the size of the difference, and then replacing the outlier with the average value of the two data points before and after it.

[0030] Preferably, the outlet pressure signal is normalized by using z-score.

[0031] The maximum value of the distance between all sub-sequences of the time sequence and the nearest neighbor sub-sequences thereof which are not self-matching is calculated by using the matrix profile, so as to measure the self-similarity inside the whole time sequence.

[0032] The plunger pump abnormality detection system based on pressure pulsation similarity provided by the application comprises:

[0033] Module M1: collecting the pressure signal of the plunger pump outlet through the high-frequency pressure sensor installed at the outlet of the plunger pump;

[0034] Module M2: judging whether the collected pressure signal has a trend item change, and if the change occurs, removing the trend item of low-frequency change by using the detrending algorithm;

[0035] Module M3: removing the abnormal points of the outlet pressure signal;

[0036] Module M4: performing Fourier transform on the signal of the pressure sensor to obtain the frequency spectrum of the pressure signal, judging whether the frequency with the highest amplitude value in the frequency spectrum is the motion base frequency or the multiple frequency of the plunger pump, if yes, preliminarily judging that the plunger pump is normal, executing modules M5-S7, otherwise, outputting the plunger pump abnormality or the failure to collect the normal plunger pump pressure pulsation signal;

[0037] Module M5: performing normalization processing on the pressure signal;

[0038] Module M6: calculating the plunger pump health factor based on the pressure pulsation similarity based on the normalized outlet pressure signal;

[0039] Module M7: judging whether the plunger pump is abnormal according to the preset threshold value based on the calculated health factor.

[0040] Preferably, the kinematic frequency of the plunger pump is calculated as follows:

[0041]

[0042] Wherein, n is the rotation speed of the plunger pump, and k is the number of plungers of the plunger pump;

[0043]

[0044] Wherein, FS is the sampling frequency;

[0045] According to the sampling frequency FS, the rotation speed of the plunger pump and the number of plungers of the plunger pump, the data point number L1 of the plunger pump outlet pressure signal segment corresponding to one kinematic frequency f under the sampling frequency FS can be calculated:

[0046]

[0047] The data points of the outlet pressure points per rotation of the plunger pump are L=L1*k;

[0048]

[0049] Preferably, the pressure signal is decomposed into k intrinsic mode functions (IMFs) and a residual term R by using a variational mode decomposition algorithm, and the expression is:

[0050] [IMFs, R] = vmd(T)

[0051] Wherein, vmd represents the variational mode decomposition, T is the outlet pressure signal, IMFs is the intrinsic mode function, and R is the residual term.

[0052] Then, the IMFs are selected to find the IMFs where the trend items are located, and then the original signal is subtracted from all trend items to obtain the detrended outlet pressure signal:

[0053]

[0054] Wherein, represents the detrended outlet pressure signal, represents the IMFs where the trend items are located.

[0055] Preferably, the outlier removal algorithm is used to remove outliers, including judging the difference between two points in the time series, and judging whether it is an outlier by the size of the difference, and then replacing the outlier with the average value of the two data points before and after.

[0056] Preferably, the outlet pressure signal is normalized by using z-score.

[0057] The maximum value of the distance between all sub-sequences of the time series and its non-self-matching nearest neighbor sub-sequences is calculated by using the matrix profile, so as to measure the self-similarity of the whole time series.

[0058] Compared with the prior art, the present application has the following beneficial effects: the present application proposes an unsupervised plunger pump anomaly monitoring method, which does not require historical fault data, and performs anomaly monitoring on the plunger pump based on the similarity of pressure pulsation, and can solve the problems of load change of hydraulic system pressure in actual engineering and difficulty in collecting a large amount of fault label data. BRIEF DESCRIPTION OF DRAWINGS

[0059] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0060] Figure 1 The flow chart of the plunger pump anomaly detection method of the present application;

[0061] Figure 2 for the outlet pressure signal with trend term;

[0062] Figure 3 for the de-trended outlet pressure signal;

[0063] Figure 4 for the outlet pressure signal and its frequency spectrum;

[0064] Figure 5 for the outlet pressure signal and its HI under healthy condition;

[0065] Figure 6 for the outlet pressure signal and its HI under single loose shoe fault;

[0066] Figure 7 for the outlet pressure signal and its HI under artificial added fault. DETAILED DESCRIPTION

[0067] The present application will be described in detail below with specific embodiments. The following examples will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present application. These are within the scope of the present application.

[0068] Embodiment:

[0069] Reference Figure 1 The present application proposes a plunger pump anomaly detection algorithm based on pressure pulsation similarity, including the following steps:

[0070] S1: A high-frequency pressure sensor is installed at the outlet of the plunger pump to collect the pressure signal at the outlet of the plunger pump;

[0071] The kinematic frequency of the plunger pump is calculated as follows:

[0072]

[0073] where n is the rotational speed of the plunger pump (unit rpm), and k is the number of plungers of the plunger pump;

[0074] The signal 10 times the kinematic frequency of the plunger pump is collected, considering the sampling theorem (collecting 2 times the highest signal frequency), and the sampling frequency of the outlet pressure signal of the plunger pump is set to be 20 times or more than the kinematic frequency of the plunger pump.

[0075]

[0076] where FS is the sampling frequency.

[0077] According to the sampling frequency FS, the plunger pump speed, the number of plunger pump plungers, the approximate data points L1 of the plunger pump outlet pressure signal segment corresponding to a kinematic frequency f under the sampling frequency FS can be calculated:

[0078]

[0079] The approximate data points of the plunger pump outlet pressure points for one rotation are L=L1*k

[0080]

[0081] For a plunger pump in a healthy state, the health states of the various parts of the pump are similar, so it is a symmetrical structure, and the outlet pressure signal is a pseudo-periodic signal, and the length of the period is approximately L1, so the time sequence subsequence with a length of L1 is called a pressure pulsation of the plunger pump outlet pressure.

[0082] S2: For the collected pressure signal, whether a trend item has changed, that is, whether the mean value of the high-frequency pressure signal has changed over time, if a change has occurred, a detrending algorithm is used to remove the low-frequency change trend item, and the detrending algorithm includes but is not limited to polynomial fitting trend, filter, time-frequency decomposition algorithm (empirical mode decomposition, variational mode decomposition) and the like to remove the signal trend item.

[0083] As shown in Figure 2 , an example of a plunger pump outlet pressure signal with a trend item change is given.

[0084] The variational mode decomposition algorithm is used to decompose the pressure signal into k intrinsic mode functions IMFs and a residual term R. Specifically:

[0085] [IMFs, R] = vmd(T)

[0086] Wherein, vmd represents variational mode decomposition, T is the outlet pressure signal, IMFs is the intrinsic mode function, and R is the residual term.

[0087] Then select IMFs to find the IMFs where the trend item is located, and then subtract all trend items from the original signal to obtain the detrended outlet pressure signal:

[0088]

[0089] Wherein, represents the detrended outlet pressure signal, represents the IMFs where the trend item is located.

[0090] The trend item calculated from the example outlet pressure signal is shown in Figure 2 , and the detrended outlet pressure signal is shown inFigure 3 As shown.

[0091] S3: Remove abnormal points from the outlet pressure signal of the plunger pump.

[0092] Anomalies include, but are not limited to, anomalies that have undergone large step changes.

[0093] The pump outlet pressure will not change rapidly. If a large step change occurs, it is an outlier caused by electromagnetic interference during signal acquisition. Outlier removal algorithms are used to remove outliers, including but not limited to, judging the difference between two points in the time series to determine whether it is an outlier by the magnitude of the difference, and then replacing the outlier with the average of the two data points before and after.

[0094] S4: Perform a Fourier transform on the plunger pump outlet pressure signal to obtain the spectrum of the pressure signal, and determine whether the spectrum of the plunger pump outlet pressure signal is normal. If it is, preliminarily determine that the plunger pump is normal, and proceed to S5~S8. If not, output that the plunger pump is abnormal or that a normal plunger pump pressure pulsation signal has not been acquired.

[0095] Determine whether the spectrum of the plunger pump outlet pressure is normal, including but not limited to whether the frequency with the highest spectrum amplitude is the kinematic fundamental frequency or its harmonics of the plunger pump.

[0096] like Figure 4 As shown, an example of a plunger pump outlet pressure signal and its spectrum is given.

[0097] The outlet pressure signal was acquired at a motor speed of 1000 rpm. This is a 9-piston pump, therefore the kinematic frequency of the pump is:

[0098]

[0099] From Figure 2 It can be seen that the highest energy in the plunger pump's spectrum are all the plunger pump's kinematic frequency and its harmonics, which is consistent with theoretical calculations. This indicates that normal pressure pulsations have been collected and the plunger pump is preliminarily judged to be normal (if other frequencies have higher amplitudes, it means that the flow pulsations caused by the fault account for a large proportion of the total flow pulsations, and the plunger pump is abnormal).

[0100] S5: Normalize the pressure signal, including but not limited to normalization using z-score:

[0101]

[0102] S6: Calculate the plunger pump health factor based on pressure pulsation similarity for the normalized outlet pressure signal.

[0103] The outlet pressure signal of the plunger pump per rotation contains the state information of the plunger pump per cycle, so the health information of the plunger pump per L data points can be determined.

[0104] The normalized outlet pressure sensor signal T=[t1, t2,...,t n ] is collected, wherein n is the length of the outlet pressure signal.

[0105] Therefore, if 1 data point is set per interval, the health factor is calculated once, so for the normalized outlet pressure signal T=[t1, t2,...,t n ], the health factor HI=[hi1, hi2,...,hi n-L ] of length n-L can be calculated.

[0106] The calculation of the single health factor hi i is obtained by calculating the T i ′=[t i ,t i+1 ,...,t i+L-1 ] signal.

[0107] In the healthy state, the health status of each part (for example: k plungers) of the plunger pump is similar, so the similarity between the k pressure pulsations in one rotation of the plunger pump is similar, and when the plunger pump is degraded, the similarity between the pressure pulsations decreases, so the similarity of the pressure pulsations in the outlet pressure signal per rotation can be used as the health factor of the plunger pump.

[0108] For a time sequence T=[t1, t2,...,t L ] of length L, the subsequence T j,k of the time sequence is defined as a time sequence of length k starting from j, T j,k =[t j ,t j+1 ,...,t j+L-1 ], 1≤j≤L-k+1.

[0109] Therefore, the subsequence of length L1 is a pressure pulsation, and there are L-L1+1 subsequences for the outlet pressure signal per rotation.

[0110] Therefore, the similarity between the L-L1+1 subsequences can be used as the health factor of the plunger pump. Before calculating the similarity between the L-L1+1 subsequences, the similarity between the self-matching subsequences needs to be excluded.

[0111] For the subsequence T j,k , the k / 2 regions before and after the subsequence are set as isolation zones, that is, the subsequence Tj-k / 2,k , T j-k / 2+1,k ,...,T j+k / 2,k is defined as the distance between T j,k and its non-self-matching nearest neighbor.

[0112] Therefore, the similarity between non-self-matching subsequences is defined as the plunger pump health factor.

[0113] Therefore, for a subsequence T j,k , the subsequence with the largest similarity to T , i.e., the subsequence with the smallest distance (the nearest neighbor subsequence), can measure the similarity of T

[0114] The distance includes but is not limited to the Euclidean distance, the dynamic programming distance, etc.

[0115] The similarity between non-self-matching subsequences includes but is not limited to the distance between the subsequence and its non-self-matching nearest neighbor subsequence.

[0116] And the self-similarity within the entire time series T includes but is not limited to the maximum value of the distance between all subsequences of T and their non-self-matching nearest neighbor subsequences.

[0117] The Matrix profile is a fast approximation algorithm for calculating the distance between a subsequence of T and its non-self-matching nearest neighbor subsequence.

[0118] For a time series T of length L and a subsequence length L1, the Matrix profile MP

[0119]

[0120] Therefore, max(MP T ) is the maximum value of the distance between a subsequence and its non-self-matching nearest neighbor subsequence.

[0121] Therefore, for the normalized outlet pressure signal T = [t1, t2,..., t n ], the plunger pump health factor of length n-L+1 can be calculated as

[0122] Figure 5 An example of the outlet pressure signal and its health factor based on the similarity of pressure pulsations under a healthy state is given. Figure 6 An example of the outlet pressure signal and its health factor based on the similarity of pressure pulsations under a single loose shoe fault is given. Figure 7 An example of the outlet pressure signal and its health factor based on the similarity of pressure pulsations under an artificial simulated fault is given. FromFigure 5 and Figure 6 It can be seen that the similarity of the plunger pump in the healthy state is high, and the loose shoe fault affects two pressure pulsation waveforms, resulting in the decline of the similarity of the non-self-matching subsequence within one rotation, so the health factor based on the pressure pulsation similarity can well divide the health state of the plunger pump, and the abnormal detection function can be easily realized by setting a threshold. Figure 7 is in the healthy state, the artificial changes the change of the pressure pulsation waveform, it can be seen that in the non-fault area, the health factor is less than the threshold, and in the artificially introduced fault interval, the proposed algorithm can also well locate the anomaly.

[0123] Those skilled in the art know that, in addition to implementing the system, device and each module thereof provided by the present application in the form of pure computer readable program code, the same program can also be realized by logically programming the method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers. Therefore, the system, device and each module thereof provided by the present application can be considered as a hardware component, and the modules included therein for realizing various programs can also be considered as structures within the hardware component; the modules for realizing various functions can also be considered as both software programs for realizing methods and structures within the hardware component.

[0124] The specific embodiments of the present application are described above. It needs to be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the present application. In the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other at will.

Claims

1. A plunger pump abnormality detection method based on pressure pulsation similarity, characterized by, Comprising: Step S1: install a high-frequency pressure sensor at the outlet of the plunger pump to collect the pressure signal at the outlet of the plunger pump; Step S2: determine whether the collected pressure signal has a trend item change, if there is a change, use the detrending algorithm to remove the low-frequency change trend item; Step S3: remove abnormal points from the outlet pressure signal; Step S4: perform Fourier transform on the signal of the pressure sensor to obtain the frequency spectrum of the pressure signal, and determine whether the frequency with the highest amplitude in the frequency spectrum is the motion base frequency or its multiple frequency of the plunger pump, if so, it is preliminarily determined that the plunger pump is normal, and steps S5-S7 are executed, otherwise, output the plunger pump abnormal or unable to collect normal plunger pump pressure pulsation signal; Step S5: normalize the pressure signal; Step S6: calculate the plunger pump health factor based on pressure pulsation similarity for the normalized outlet pressure signal; Step S7: determine whether the plunger pump is abnormal according to the preset threshold value according to the calculated health factor.

2. The pressure pulsation similarity-based plunger pump abnormality detection method according to claim 1, characterized by, The kinematic frequency of the plunger pump is calculated as follows: Wherein, n is the rotating speed of the plunger pump, k is the number of plungers of the plunger pump; Wherein, FS is the sampling frequency; According to the sampling frequency FS, the rotating speed of the plunger pump and the number of plungers of the plunger pump, the data point number L1 of the plunger pump outlet pressure signal segment corresponding to one kinematic frequency f at the sampling frequency FS can be calculated: The data point number of the outlet pressure point of the plunger pump rotating one round is L=L1*k; 3. The pressure pulsation similarity-based plunger pump anomaly detection method according to claim 1, characterized by, The pressure signal is decomposed by using the variational mode decomposition algorithm, and is decomposed into k intrinsic mode functions IMFs and a residual term R, and the expression is: [IMFs, R] = vmd(T) Wherein, vmd represents variational mode decomposition, T is the outlet pressure signal, IMFs is the intrinsic mode function, and R is the residual term; Then select IMFs to find the IMFs where the trend item is, and then subtract all trend items from the original signal to obtain the detrended outlet pressure signal: wherein, represents the detrended outlet pressure signal, denotes the IMFs in which the trend component is located.

4. The pressure pulsation similarity-based plunger pump abnormality detection method according to claim 1, characterized by, Anomaly value removal algorithm is used to remove abnormal values, including judging the difference between two points in time sequence, and judging whether it is an abnormal value by the size of the difference, and then replacing the abnormal value with the average value of the two data points before and after it.

5. The pressure pulsation similarity-based plunger pump anomaly detection method according to claim 1, characterized by, The outlet pressure signal is normalized by using z-score. The maximum value of the distance between all sub-sequences of the time sequence and its non-self-matching nearest neighbor sub-sequences is calculated by using matrix profile, so as to measure the self-similarity of the whole time sequence.

6. A plunger pump anomaly detection system based on pressure pulsation similarity, characterized by, Comprising: Module M1: collect the pressure signal at the outlet of the plunger pump by installing a high-frequency pressure sensor at the outlet of the plunger pump; Module M2: determine whether the collected pressure signal has a trend item change, if there is a change, use the detrending algorithm to remove the low-frequency change trend item; Module M3: remove abnormal points from the outlet pressure signal; Module M4: perform Fourier transform on the signal of the pressure sensor to obtain the frequency spectrum of the pressure signal, and determine whether the frequency with the highest amplitude in the frequency spectrum is the motion base frequency or its multiple frequency of the plunger pump, if so, it is preliminarily determined that the plunger pump is normal, and modules M5-S7 are executed, otherwise, output the plunger pump abnormal or unable to collect normal plunger pump pressure pulsation signal; Module M5: normalize the pressure signal; Module M6: calculating the plunger pump health factor based on pressure pulsation similarity on the normalized outlet pressure signal; Module M7: judging whether the plunger pump is abnormal according to the preset threshold value according to the calculated health factor.

7. The pressure pulsation similarity based plunger pump anomaly detection system of claim 6, wherein, The kinematic frequency of the plunger pump is calculated as follows: Wherein, n is the rotation speed of the plunger pump, and k is the number of plungers of the plunger pump; Wherein, FS is the sampling frequency; According to the sampling frequency FS, the rotation speed of the plunger pump and the number of plungers of the plunger pump, the data point number L1 of the outlet pressure signal segment of the plunger pump corresponding to one kinematic frequency f under the sampling frequency FS can be calculated: The data point number of the outlet pressure points of the plunger pump rotating one round is L=L1*k; 8. The pressure pulsation similarity based plunger pump anomaly detection system of claim 6, wherein, The pressure signal is decomposed into k intrinsic mode functions IMFs and a residual term R by using the variational mode decomposition algorithm, and the expression is: [IMFs, R] = vmd(T) Wherein, vmd represents the variational mode decomposition, T is the outlet pressure signal, IMFs is the intrinsic mode function, and R is the residual term; Then, the IMFs are selected to find the IMFs where the trend items are located, and then the original signal is subtracted from all the trend items to obtain the detrended outlet pressure signal: wherein, represents the detrended outlet pressure signal, represents the IMFs in which the trend component is located.

9. The pressure pulsation similarity based plunger pump anomaly detection system of claim 6, wherein, An outlier removal algorithm is used to remove outliers, including judging the difference between two points in a time series, judging whether it is an outlier by the size of the difference, and then replacing the outlier with the average value of the two data points before and after it.

10. The pressure pulsation similarity based plunger pump anomaly detection system of claim 6, wherein, The outlet pressure signal is normalized by using z-score. The maximum value of the distance between all sub-sequences of the time series and their non-self-matching nearest neighbor sub-sequences is calculated by using matrix profile, so as to measure the self-similarity inside the whole time series.

Images